Reply to thread

Message

<blockquote data-quote="Andy Ful" data-source="post: 1023158" data-attributes="member: 32260">Does Population Size matter?When calculating the exact value of the sample size &quot;s&quot; we need to know the Population Size &quot;N&quot;. But, it is worth knowing that in many cases the sample size does not significantly differ from its value &quot;s0&quot; calculated for unlimited Population Size. The same is true for the Margin of Error &quot;e&quot; which is usually almost the same as its value &quot;e0&quot; calculated for unlimited Population Size.We have:If&nbsp; s/N = d &lt;&lt; 1 Then:e ~ e0*(1 - d/2)s ~ s0*(1-d)A typical situation in any R-W test is that s &gt;= 100 and d &lt; 1/100.To show this, I took the data from the SonicWall report:[ATTACH=full]272653[/ATTACH]If we use the data from the year 2021, then we can calculate over 1200 samples per 1 day. In a typical R-W test, approximately 6 samples per 1 day are used. In such a case we have:d ~ 6/1200 = 0.005We can use the unlimited Population Size (N ----&gt; infinity) and the difference in results will be:Sample Size (with a relative error of 0.5%)Margin of Error (with a relative error of 0.25%)In reality, these errors should be even smaller, because in tests we can have also some samples that are not Never-Before-Seen.Post updated/corrected.</blockquote>

[QUOTE="Andy Ful, post: 1023158, member: 32260"] [B]Does Population Size matter?[/B] When calculating the exact value of the sample size "s" we need to know the Population Size "N". But, it is worth knowing that in many cases the sample size does not significantly differ from its value "s0" calculated for unlimited Population Size. The same is true for the Margin of Error "e" which is usually almost the same as its value "e0" calculated for unlimited Population Size. We have: If s/N = d << 1 Then: e ~ e0*(1 - d/2) s ~ s0*(1-d) A typical situation in any R-W test is that s >= 100 and d < 1/100. To show this, I took the data from the SonicWall report: [ATTACH type="full" alt="1675540422097.png"]272653[/ATTACH] If we use the data from the year 2021, then we can calculate over 1200 samples per 1 day. In a typical R-W test, approximately 6 samples per 1 day are used. In such a case we have: d ~ 6/1200 = 0.005 We can use the unlimited Population Size (N ----> infinity) and the difference in results will be: Sample Size (with a relative error of 0.5%) Margin of Error (with a relative error of 0.25%) In reality, these errors should be even smaller, because in tests we can have also some samples that are not Never-Before-Seen. Post updated/corrected. [/QUOTE]

Verification